Methodology

Evidence before confidence.

Cortex combines AI-assisted research and reasoning with retained external evidence. It is designed to make strong claims only when the evidence justifies them—and to expose limitations when it does not.

The evidence path

From a question to an accountable report

Cortex does not ask models to vote on truth. It creates a bounded research record, qualifies retained evidence, examines how that evidence relates to material claims, and reports what the evidence can—and cannot—support.

  1. 01

    Question or claim

    ASK begins with a question. CHECK begins with a supplied claim or document.

  2. 02

    Research

    Cortex performs bounded retrieval appropriate to the selected workflow.

  3. 03

    Evidence qualification

    Sources are assessed for relevance, authority, methodology, directness, independence, and timeliness where applicable.

  4. 04

    Contradiction analysis

    Support, qualifications, scope differences, uncertainty, and substantive contradictions are kept distinct.

  5. 05

    Adjudication

    Material evidence is reconciled against the central factual proposition and report claims.

  6. 06

    Trust and limitations

    Cortex reports evidence-derived diagnostics, sources, and meaningful gaps—or Insufficient Evidence.

ASK and CHECK solve different problems

ASK

“I have a question. Research it.”

ASK researches a question and produces an evidence-backed answer. Use it to investigate a topic, compare claims, or understand what retained evidence supports.

Start ASK →

CHECK

“I have a claim or document. Evaluate it.”

CHECK evaluates supplied claims or documents against retained external evidence. It is one Standard verification workflow; documents up to 2,500 words use 1 credit and documents from 2,501–8,000 words use 2.

Start CHECK →

Choose the right ASK depth

More depth means more extensive bounded investigation—not guaranteed correctness.

Snapshot

Quick orientation

Five promotional uses per day. No Cortex Analysis Credit. Snapshot does not include Canonical Trust.

Standard

Normal evidence-backed research

Use for most research questions. 1 Cortex Analysis Credit.

Deep

More extensive investigation

Use for complex or consequential questions that deserve deeper bounded research and analysis. 3 Cortex Analysis Credits.

Evidence Responsibility

Claims should not outrun their evidence.

Evidence Responsibility is Cortex's commitment to connect material claims to retained evidence, distinguish evidence quality from fluent model output, and disclose meaningful limits. It is a product principle—not a certification or legal standard.

Core principle: AI output is not evidence.

What Cortex Trust means

Cortex Trust is a 0–100 evidence-derived assessment of a central factual proposition. It reflects the retained evidence record; it is not a literal probability that an answer is true.

Cortex Trust is not AI confidence, model confidence, model agreement, majority voting, or a substitute for professional judgment.

Reports may also show Research Confidence (evidence coverage and sufficiency), Report Integrity (citation, provenance, and report quality), and Model Agreement (alignment among model outputs). These diagnostics answer different questions. Model Agreement is diagnostic only; it is not evidence.

Evidence outcomes

SupportedRetained evidence supports the proposition under the relevant scope.
Qualified / Partially SupportedThe core proposition has material limits, conditions, or narrower applicability.
ContestedCredible retained evidence materially differs and the disagreement remains relevant.
ContradictedRetained evidence materially opposes the proposition under matching scope.
Unverifiable / Insufficient EvidenceThe available qualified evidence did not establish a stronger conclusion.

Why “Insufficient Evidence” matters

When Cortex cannot retain enough qualified evidence to justify a stronger conclusion, it can withhold a numeric Trust result and report Insufficient Evidence. This does not mean the proposition is false. It means the available evidence in that run did not support a stronger evidentiary conclusion.

A trustworthy system must be able to say: the evidence is not sufficient.

Contradiction is not the same as difference

Cortex distinguishes, where the evidence permits, between substantive contradiction, narrower scope, qualification, methodological difference, interpretive difference, uncertainty, and insufficient evidence. A source that addresses a narrower population, jurisdiction, timeframe, or boundary condition does not automatically contradict a broader claim. Cortex does not promise that every contradiction can be resolved.

Limitations

  • Important evidence may be unavailable, private, paywalled, unpublished, or absent from the public web.
  • Current events, law, medicine, markets, and other time-sensitive subjects can change after a report is produced.
  • Sources vary in authority, methodology, independence, directness, and relevance to the exact proposition.
  • Ambiguous or compound claims may require clarification before they can be assessed cleanly.
  • Retrieval and provider failures can reduce evidence coverage; Cortex may fail closed rather than overstate a conclusion.
  • AI-assisted reasoning and synthesis can make mistakes. Models help analyze; they do not become evidence.
  • Cortex supports professional judgment but does not replace qualified medical, legal, financial, regulatory, or other specialist advice.

Choose the workflow that matches your task

Research a question with ASK, evaluate an existing claim or document with CHECK, or continue through the Resources guide.

Use Cortex →Explore resources